If you run your playbook with ansible-playbook -vvv
you'll see the actual command being run, so you can check whether the key is actually being included in the ssh command (and you might discover that the problem was the wrong username rather than the missing key).
I agree with Brian's comment above (and zigam's edit) that the vars section is too late. I also tested including the key in the on-the-fly definition of the host like this
# fails
- name: Add all instance public IPs to host group
add_host: hostname={{ item.public_ip }} groups=ec2hosts ansible_ssh_private_key_file=~/.aws/dev_staging.pem
loop: "{{ ec2.instances }}"
but that fails too.
So this is not an answer. Just some debugging help and things not to try.